Content area
Functional Reactive Programming (FRP) is a promising class of abstractions for interactive programs. FRP systems provide values defined at all points in time (behaviors or signals) and values defined at countably many points in time (events) as abstractions.
Signal-function FRP is a subclass of FRP which does not provide direct access to time-varying values to the programmer, but instead provides signal functions, which are reactive transformers of signals and events, as first-class objects in the program.
All signal-function implementations of FRP to date have utilized demand-driven or "pull-based" evaluation for both events and signals, producing output from the FRP system whenever the consumer of the output is ready. This greatly simplifies the implementation of signal-function FRP systems, but leads to inefficient and wasteful evaluation of the FRP system when this strategy is employed to evaluate events, because the components of the signal function which process events must be computed whether or not there is an event occurrence.
In contrast, an input-driven or "push-based" system evaluates the network whenever new input is available. This frees the system from evaluating the network when nothing has changed, and then only the components necessary to react to the input are re-evaluated. This form of evaluation has been applied to events in standard FRP systems but not in signal-function FRP systems.
I describe the design and implementation of a signal-function FRP system which applies pull-based evaluation to signals and push-based evaluation to events (a "push-pull" system). The semantics of the system are discussed, and its performance and expressiveness for practical examples of interactive programs are compared to existing signal-function FRP systems through the implementation of a networking application.